ReBeCA: Unveiling Interpretable Behavior Hierarchy behind the Iterative Self-Reflection of Language Models with Causal Analysis

2026-02-06

AI Summary

This work addresses the limited interpretability and poor generalizability of existing self-reflection mechanisms in language models. The authors propose a causal analysis–based approach to model self-reflection trajectories, employing a three-stage Invariant Causal Prediction (ICP) pipeline to construct causal graphs that identify the true semantic behaviors—and their hierarchical structure—underlying self-reflection outcomes. Their analysis reveals, for the first time, that only a sparse subset of semantic behaviors exhibits generalization capability, and that the concurrent presence of multiple positive behaviors can paradoxically diminish overall effectiveness. Experimental results demonstrate that the identified sparse set of causal parent nodes yields significantly improved stability on out-of-distribution data (p = .013, η²ₚ = .071), with a 49.6% increase in structural likelihood.

Technology Category

• Machine Learning: Causal Learning
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Reasoning under Uncertainty: Causality

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

While self-reflection can enhance language model reliability, its underlying mechanisms remain opaque, with existing analyses often yielding correlation-based insights that fail to generalize. To address this, we introduce \textbf{\texttt{ReBeCA}} (self-\textbf{\texttt{Re}}flection \textbf{\texttt{Be}}havior explained through \textbf{\texttt{C}}ausal \textbf{\texttt{A}}nalysis), a framework that unveils the interpretable behavioral hierarchy governing the self-reflection outcome. By modeling self-reflection trajectories as causal graphs, ReBeCA isolates genuine determinants of performance through a three-stage Invariant Causal Prediction (ICP) pipeline. We establish three critical findings: (1) \textbf{Behavioral hierarchy:} Semantic behaviors of the model influence final self-reflection results hierarchically: directly or indirectly; (2) \textbf{Causation matters:} Generalizability in self-reflection effects is limited to just a few semantic behaviors; (3) \textbf{More $\mathbf{\neq}$ better:} The confluence of seemingly positive semantic behaviors, even among direct causal factors, can impair the efficacy of self-reflection. ICP-based verification identifies sparse causal parents achieving up to $49.6\%$ structural likelihood gains, stable across tasks where correlation-based patterns fail. Intervention studies on novel datasets confirm these causal relationships hold out-of-distribution ($p = .013, \eta^2_\mathrm{p} = .071$). ReBeCA thus provides a rigorous methodology for disentangling genuine causal mechanisms from spurious associations in self-reflection dynamics.

Problem

• self-reflection
• causal analysis
• behavior hierarchy
• language models
• generalizability

Innovation

• causal analysis
• self-reflection
• behavior hierarchy
• invariant causal prediction
• language models